BrainSelector chooses a model, reasoning level and speed for each Codex task.
Install at least one Codex client: the ChatGPT app, the VS Code Codex extension, or the standalone Codex CLI.
Open BrainSelector once to connect the standalone Codex CLI automatically. Start a new terminal and run codex to use BrainSelector; existing sessions can stay open and finish their work. brainselector tui opens the BrainSelector dashboard.
The app is available for Mac, Linux, and as a VS Code extension. You can install just one of them, then the dashboard will allow you to install the other ones if needed. Adding another interface on the same computer reuses your history and settings.
| Start with | Best for | Platform |
|---|---|---|
| Mac app | ChatGPT desktop app with a menu-bar dashboard | Apple Silicon, macOS 15 or later |
| Linux app | A dashboard in your browser, opened from the applications menu | Linux x86_64; RHEL 9.8 build baseline |
| VS Code Marketplace | Codex or Claude in VS Code, including Remote-SSH | The same Mac and Linux platforms |
| VSIX download | Offline/manual VS Code installation | The same Mac and Linux platforms |
Download the latest release and view checksums.
On Mac, unpack the signed download and move BrainSelector to Applications, or use Homebrew:
brew install --cask jniedzie/brainselector/brainselectorThe Mac app is Developer ID signed and notarized. Python and the provider helper are included; no separate runtime installation is needed. Release status and checksums.
The full app is not available in the Mac App Store because its desktop routing, shared backend and device installer are incompatible with App Sandbox requirements. A separate companion edition is planned.
The Linux desktop-client routing is very experimental (=untested).
Open BrainSelector once after installation. If ChatGPT or VS Code was already running, use the restart or reload action shown by BrainSelector. Existing terminal sessions can finish normally; start a new codex session to use routing.
To see more details:
- Open BrainSelector using the brain icon in the macOS menu bar, the VS Code extension view, or
brainselector tui. - Have a look at all the stats in the dashboard.
- Check the Settings section for more control over BrainSelector.
- Review task history, rate advisor decisions, optionally provide a comment, then save it.
You can pause routing in the dashboard to keep your selected model for new tasks. For one task, include \skip - it will be removed from the prompt and preserve your model selection for that prompt. Code examples and longer names are left unchanged. The previous commands still work.
Use Install on your devices to add another interface or install the extension on a configured SSH host. SSH sign-in must already work; remote extension installation requires an existing VS Code server. Linked devices exchange history and settings when connected.
Savings are estimates using recorded tokens and model rates, but not your actual bill - take it with a large grain of salt. Advisor calls also consume usage and appear separately.
Your local history contains task text and ratings. Routing sends task text and, when needed, recent context to your OpenAI advisor through your existing Codex sign-in. Linked SSH devices exchange execution metadata, task labels, ratings, notes and routing settings; full prompts stay on their original installation.
Community sharing is enabled by default. Clients send category, model, rating and usage statistics automatically. They do not upload prompts, titles, notes, task IDs, host names or OpenAI credentials. Turn sharing off in Settings to stop future uploads.
Report a problem. Include the BrainSelector version, platform, what you expected and what happened. Reports are public: remove private task text and credentials.
BrainSelector covers supported native Codex and Claude Code tasks, not ordinary ChatGPT web conversations. Native client updates can affect compatibility. Quitting the dashboard does not disable routing; pause it first. Uninstall BrainSelector through VS Code’s Extensions view. Its interface and executable binding are cleaned up automatically; finish active tasks before accepting a reload. Feedback, usage history, shared settings, and the shared engine are retained for other interfaces and later reinstalls. A small compatibility launcher on remote hosts lets Codex start normally when your laptop still uses BrainSelector.